StatefulSet vs Deployment

Hands-on example

Assume:

• A working Kubernetes cluster
• kubectl is configured

---

Initial state

No workloads exist.

kubectl get deployments
kubectl get statefulsets
kubectl get pods

No resources found in default namespace.
No resources found in default namespace.
No resources found in default namespace.

---

Step 1: Create a Deployment

kubectl create deployment web --image=nginx --replicas=2

deployment.apps/web created

Check Pods.

kubectl get pods

web-6f7c8d9f6b-abcde   1/1   Running   0   5s
web-6f7c8d9f6b-fghij   1/1   Running   0   5s

Delete one Pod.

kubectl delete pod web-6f7c8d9f6b-abcde

pod "web-6f7c8d9f6b-abcde" deleted

Check again.

kubectl get pods

web-6f7c8d9f6b-klmno   1/1   Running   0   5s
web-6f7c8d9f6b-fghij   1/1   Running   0   20s

What changed:

• Pod name changed
• A new Pod replaced the deleted one

What did not change:

• No identity was preserved

---

Step 2: Create a StatefulSet

Create a headless Service first.

kubectl create service clusterip db --tcp=80:80 --dry-run=client -o yaml | sed 's/ClusterIP/None/' | kubectl apply -f -

service/db created

Create the StatefulSet.

kubectl create statefulset db --image=nginx --replicas=2 --service-name=db

statefulset.apps/db created

Check Pods.

kubectl get pods

db-0   1/1   Running   0   5s
db-1   1/1   Running   0   3s

Delete db-0.

kubectl delete pod db-0

pod "db-0" deleted

Check again.

kubectl get pods

db-0   1/1   Running   0   5s
db-1   1/1   Running   0   30s

What changed:

• db-0 was recreated

What did not change:

• Pod name and identity stayed the same

---

Key observation

• Use Deployments for stateless workloads
• Use StatefulSets when Pod identity must be stable